Explainable radiomics based on association of histopathological cell density and multiparametric MR radiomic features for high-risk stratification of prostate cancer patients.

Objective: This study aimed to develop an explainable radiomics model for stratifying prostate cancer (PCa) patients with high-risk disease via investigation of the association between cell density (CD) in the PCa region on histopathological images and multiparametric MR (mpMR) radiomics features.

Materials and methods: 137,970 radiomic features were calculated from mpMR images (101 PCa regions of 44 patients), and joint histograms (JHs) were derived from dynamic contrast-enhanced (DCE) images for each PCa region. The association between CD on histopathological images and its corresponding mpMR radiomic features in PCa regions for various grade groups and the three risk groups was evaluated using Spearman's correlation coefficient. To validate the potential of the radiomic-feature-CD association, we developed the radiomics model for stratifying patients into low/intermediate-risk and high-risk groups.

Results: There were moderate correlations of the CD with a DCE-based texture feature (WV_HH_1st_GLSZM_ZP) (ρ = 0.609, p = 0.024) and DCE-JH feature (JH_WV_HL_1st versus 5th-1st_Hist_STD) (ρ = 0.609, p = 0.024) in the high-risk group. The radiomics model had an accuracy of 0.920 for stratifying the patients of a test dataset into the low/intermediate-risk and high-risk groups.

Conclusion: The association between CD and mpMR features can be leveraged to develop the explainable radiomics for the high-risk stratification of patients with PCa.